Richard Bucker

Performance Testing BSD Linux

Posted at — Jan 7, 2022

I did not start my day expecting to do some performance testing. Instead I started deploying updated versions of SQLite(3.37.0200) and golang(1.17.6). I have a number of Intel NUCs running different flavors of OS. I started my builds and I found myself wondering why OpenBSD 7 was taking so long to compile one project (golang)… while, in other terminal windows, FreeBSD 13 had completed the task.

After a quick google search I found this article. I quickly realized it was current and covered the same OS, including ClearLinux, FreeBSD13, OpenBSD7, which I use today. The author ran a number of different tests but also tests that might be considered overlapping. for example a GIT test should have been covered by the memory, file or disk tests. Sure there is something to be said for a combination test that has elements of integration latency but I don’t think that was the mission. One critical missing element was a conclusion.

should performance matter anymore?

So let me offer my conclusion. ClearLinux was a clear performance winner and the reson is obvious. ClearLinux is produced my Intel and they have a vested interest in promoting their CPUs and performance tuning at all cost is that game. One serious downside, however, is that I recently read that ClearLinux declares itself as experimental and not for production. So maybe it’s not so ideal.

I suppose other Linux distros would be just fine for most workloads and likely in step with ClearLinux. For example POP_OS is a fun choice except they do not have a server version. They’re a layer on top of Ubuntu. So that makes a nice desktop but you get the cruft.

I recently started noticing that neofetch on my FreeBSD machines ran so much faster than OpenBSD. This was at the core of my decision to check the performance. And another article I read talked about BSD bugs and the BSD development teams. On that article’s point that the OpenBSD team was fast to react to bugs, concerned about correctness, security, etc… and that FreeBSD was bloating like Linux and developer interaction was poor and slow to react to bug reports. OpenBSD was ever so slightly in the lead.

Compile time aside OpenBSD is full featured. It can virtualise with it’s own implementaion of VMM and it can encrypt the boot drive. It also does everything else that a good OS does and offers some desktop features. OpenBSD will never be windows or mac competitor on the desktop but it can hold it’s own in the server room. I’m also paranoid. I like being able to think about the OS are mostly separate from the application. The the OS is a shim to the outside world and not tightly coupled. It takes that kind of dedication to keep is separate. (and I can always buy more hardware)

So even though it seems to be the slowest it’s also the safest. Safety or performance… that is the question.